Research And Implementation Of Railway Personnel And Foreign Object Detection Algorithm For Public Security

China’s public infrastructure construction has achieved world-renowned achievements.Among them,railway transportation construction has achieved outstanding achievements.Railway travel has become the choice of more and more people,bringing great convenience to people’s production and life,but also brings some public security risks.At present,the investigation of railway security problems still mainly adopts manual inspections,which is not only inefficient,but also poses a greater security threat to the maintenance personnel.In order to investigate railway safety problems and better guarantee the safety of railway personnel,this article provides two solutions: one is to automatically obtain the position of railway personnel through surveillance video and feed it back to the railway dispatching system to generate hazard warnings in time;The second is to conduct online analysis and detection of railway safety issues such as equipment failures and foreign body intrusion through monitoring video,reducing the time for railway personnel to work.Among them,the problem of foreign object intrusion is a difficult point in the railway safety problem,due to the difficulty in defining foreign object objects and the lack of foreign object intrusion data,most supervised algorithms cannot be applied,and unsupervised methods usually fail to meet the requirements in terms of performance and stability.Based on deep learning technology,this paper conducts algorithm research and design for railway security problems,and deploys them to the railway detection system.The main research contents of this paper are as follows:(1)According to the detection requirements of railway maintenance personnel and equipment along the line,the railway object detection algorithm design is completed.YOLO algorithm is adopted as the baseline,and then four improvement measures are proposed to enhance the performance of the model,and the hyperparameters are selected through a series of comparative experiments,which realizes the efficient detection of railway personnel and equipment along the railway.(2)According to the railway foreign object intrusion detection requirements,the railway foreign object intrusion detection algorithm design is completed.First,the requirements are divided into two parts: rail area division and foreign object intrusion detection.The improved model based on U-net is used to divide the rail area.Foreign object intrusion detection uses a method based on differential and semantic information fusion proposed in this paper,and is implemented using the BU-net model proposed in this paper,the foreign object detection results are directly output after inputting the RGB picture and the differential result picture at the same time,and the spatio-temporal data augumentation method is used to reduce the scene dependence characteristics of the algorithm.The experimental results show that the foreign object intrusion detection algorithm proposed in this paper has good performance both in accuracy and generalization.(3)Deploy the railway object detection algorithm and foreign object intrusion detection algorithm to the railway detection system.Based on the Django framework,combined with Celery and Redis technologies,deployed the Tensor Flow model to the railway detection system is realized.The results of the system show that the object detection algorithm and foreign object intrusion detection algorithm in this paper have achieved the expected requirements,and provide a very valuable solution for railway safety problems.